Starstruk Logo MarkStarstruk

v0.3.65

Comfy-Org/ComfyUIv0.3.65Oct 14, 2025by github-actions[bot]

AI Summary

This release heavily prioritizes the migration to the V3 schema, converting core nodes like Flux, SD3, and LoRA extractors. It also includes updates to API nodes (Gemini, Pika) and performance enhancements for AMD hardware.

Key Highlights

  • Massive V3 schema migration for Flux, SD3, and various other nodes.
  • Refactoring of model sampling sigmas code.
  • API nodes: lazy formatting in logging and new Pika/Kling features.
  • AMD performance improvements including RDNA4 pytorch attention.

New Features

  • Enum classes for Combo options in V3-io.
  • GeminiImage-ApiNode aspect ratio and release version support.
  • Price extractor feature for API nodes.
  • Mmaudio 16k VAE implementation.

Full Release Notes

## What's Changed
* feat(V3-io): allow Enum classes for Combo options by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10237
* Refactor model sampling sigmas code. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10250
* Mvly/node update by @Kosinkadink in https://github.com/comfyanonymous/ComfyUI/pull/10042
* feat(api-nodes, pylint): use lazy formatting in logging functions by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10248
* [V3] convert nodes_model_downscale.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10199
* [V3] convert nodes_lora_extract.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10182
* [V3] convert nodes_compositing.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10174
* [V3] convert nodes_latent.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10160
* More surgical fix for #10267 by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10276
* fix(v3,api-nodes): V3 schema typing; corrected Pika API nodes by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10265
* [V3] convert nodes_sd3.py and nodes_slg.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10162
* Fix bug with applying loras on fp8 scaled without fp8 ops. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10279
* [V3] convert nodes_flux to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10122
* [V3] convert nodes_upscale_model.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10149
* Fix save audio nodes saving mono audio as stereo. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10289
* feat(GeminiImage-ApiNode): add aspect_ratio and release version of model by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10255
* feat(api-nodes): add price extractor feature; small fixes to Kling/Pika by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10284
* Update template to 0.1.95 by @comfyui-wiki in https://github.com/comfyanonymous/ComfyUI/pull/10294
* Implement the mmaudio 16k VAE. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10300
* Improve AMD performance. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10302
* Update node docs to 0.3.0 by @comfyui-wiki in https://github.com/comfyanonymous/ComfyUI/pull/10318
* update extra models paths example "clip" -> "text_encoders" by @christian-byrne in https://github.com/comfyanonymous/ComfyUI/pull/10316
* Update the extra_model_paths.yaml.example by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10319
* Always set diffusion model to eval() mode. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10331
* add indent=4 kwarg to json.dumps() in "Preview Any" node by @norgeous in https://github.com/comfyanonymous/ComfyUI/pull/10307
* WAN2.2: Fix cache VRAM leak on error by @rattus128 in https://github.com/comfyanonymous/ComfyUI/pull/10308
* [V3] convert nodes_hunyuan.py to V3 schema by @bigcat88 in https://github.com/comfyanonymous/ComfyUI/pull/10136
* Enable RDNA4 pytorch attention on ROCm 7.0 and up. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10332
* Fix loading old stable diffusion ckpt files on newer numpy. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10333
* Better memory estimation for the SD/Flux VAE on AMD. by @comfyanonymous in https://github.com/comfyanonymous/ComfyUI/pull/10334

## New Contributors
* @norgeous made their first contribution in https://github.com/comfyanonymous/ComfyUI/pull/10307

**Full Changelog**: https://github.com/comfyanonymous/ComfyUI/compare/v0.3.64...v0.3.65